• Version Mismatch: We faced the compatibility issues between different versions of ‘mmcvfull’ and ‘pytorch’ when not using the Docker image.
• Overfitting and Generalization: The model must be tuned to avoid overfitting on training data and to generalize well to new, unseen sign language gestures.
• Memory Size Limitation: The laptop’s GPU memory is insufficient for running the training with the default batch size (16). Therefore, the batch size was reduced to 2 in the config file to avoid memory errors.
• Inference Memory Error: Running the inference script for the I3D pretrained model also resulted in GPU memory errors, due to the memory intensive nature of the model.

---